[page 157↓]

Chapter 8 
General discussion

The present work examined the joint impact of environmental seasonality and controlled breeding on biological productivity of pastoral goat herds. To this end, a field experiment was conducted over a period of three years under simulated pastoral management conditions on a semi-arid thornbush savannah in northern Kenya. The experiment provided data on relevant biological performance traits of seventeen breeding groups for a total of 381 exposures distributed among six consecutive mating seasons each of two months duration, so as to achieve year-round mating, kidding, and weaning. In terms of both its design and degree of detail of data collected, the present experiment differed markedly from those conducted previously on the same subject. For instance, a total of 8547 recordings were made on survival, liveweight, and milk production of does, while survival and liveweight development of youngstock were measured at a total of 9837 time points. Also, recordings extended over entire production cycles until youngstock had reached at least the yearling stage.

Most previously published studies did not involve the experimental imposition of restricted breeding as an external treatment factor on experimental animals, but merely made use of data gathered for a wide variety of other purposes to retrospectively investigate the effects of environmental seasonality on biological performance. To this end, observations on individual animals were often grouped into several breeding or kidding seasons prior to statistical analysis. Examples of this approach with respect to goat production include: the studies on milk yields by Kennedy et al. (1980), Ruvuna et al. (1995), and Montaldo et al. (1997); the studies on reproductive performance conducted by Wilson et al. (1984), Amoah et al. (1995), Mellado et al. (1996), Ndlovu and Simela (1996); the reports on liveweight development by Wilson et al. (1984), and Ruvuna et al. (1991). To the best of our knowledge, in the context of African pastoral husbandry systems, the present study is the first attempt to investigate the biological consequences of confining breeding in goat herds to a short period in a year through the implementation of a planned and systematic breeding programme. For obvious reasons, in order to answer such complex questions as the biological consequences of seasonality, experiments are generally to be preferred to non-experimental or observational approaches in which the investigator cannot apply treatments to the animals under study and has limited control over data collection. The main problem associated with the latter approach is that causal explanations become problematic and are generally less convincing than in the former approach, which allows conclusive statements to be made on the sampled population (Hurlbert, 1984; Jager and Looman, 1995). Additionally, unplanned analyses of differences or relations perceived in the data have detective value only in the sense that the scope of inference is limited to the target population actually studied, but cannot be extended to the population at large, or inference space, implicit in the hypothesis being tested. With respect to the literature reviewed in the course of this work, these kinds of problems are most evident in relation to published results on the effects of environmental seasonality on the reproductive performance of goats. Due to their observational design many previous studies were unable to measure component traits of reproduction such as conception and prolificacy rate, and instead assessed reproductive performance in goats in terms of compound indices. This type of approach may obscure real explanations for detected differences in measured or derived variables, because underlying factors have not been measured. Compelling evidence for the presence of such confounding comes from the study of Wilson et al. (1984, 1985), who found individual flock ownership to have a large impact upon reproductive performance and overall herd productivity of Maasai goat herds.

Despite the clear-cut treatment structure adopted in this study, statistical analysis of the data and interpretation of results was complicated due to deficiencies in the design structure of the experiment. These were mainly caused by resource constraints, particularly with respect to the availability and homogeneity of experimental units, i.e., breeding does. For example, it would have been beneficial to achieve a more homogeneous repartition of does across liveweight and parity classes in each of the breeding groups. This would have facilitated studying differential effects of parity within liveweight classes on various traits, particularly on those pertaining to reproductive performance. Also, the sensitivity of the analysis could have been increased by assigning a maximum number of animals repeatedly to the same mating season over the three consecutive reproductive cycles. The problems caused by resource constraints were further exacerbated by the longitudinal nature of the study, in that experimental animals were lost to follow-up due to mortality or other reasons, thus creating additional imbalances in the design structure of the experiment. This, however, is a feature common to many longitudinal studies. Typically, individual animals are observed a different [page 158↓]number of times, at different periods of time, and the intervals between observations may be different as well. On the other hand, all production traits can be considered to be measures of performance over time and, although cross-sectional approaches are conceivable, the most reliable estimates of performance traits are generally obtained by precisely adopting a longitudinal or repeated measures type of study design.

In this regard, the general and generalized mixed model approaches used in analysing all traits, except for kid and doe survival, proved to be particularly advantageous. Firstly, it allowed the taking into account of the between-subject variability, of variability among production cycles, or both. For most traits analysed, at least one of these random effects were found to account for a significant part of the observed variation, and the analyses could be made more sensitive by removing them from the error term. Secondly, a method of analysis which allows modeling more than one error term was required because repeated measures on the same experimental animals are not independent, and cannot be analysed as if they were. The mixed model approach is a particularly effective technique to control the variability among responses from different experimental units and at the same time to account for the positive correlation between any two measurements on the same unit. A direct consequence is that the linear mixed-effects model leads to more precise estimates of outcomes for factors of interest, while it readily accommodates for the high unbalancedness typical of longitudinal data. The most commonly used method of analysis, which involves carrying out an ANOVA of the data at each time point separately (i.e., the so-called time-by-time ANOVA), provides no tests of the change of treatment effects with time and thus are generally less powerfull relative to the mixed model approach. Since treatment effects tend to vary continuously over time, quoting the time at which an effect becomes significant has very little relevance when using time-by-time ANOVA (Rowell and Walters, 1976). In the last decade, significant progress has been made in the development of methods appropriate for the analysis of longitudinal data. As these methods are now widely accessible to applied scientists, using suboptimal techniques like time-by-time ANOVA would seem unacceptable (von Ende, 1993).

In much the same way, analysing non-normally distributed outcomes such as survival rates with conventional ANOVA or linear regression methods should be avoided. With respect to survival data, the present work has shown that standard logistic regression is a particularly flexible and easy to use method of analysis which can be employed to model hazard functions parametrically over time. The proposed method provides insight into patterns of mortality over time and allows comparison among any number of risk groups simultaneously at any time point. Typical features of time-to-event data such as censoring, changes in the risk set over time, and time dependent covariables can be accommodate for, which is not possible with conventional approaches usually adopted in livestock-related research.

The results of the statistical analysis indicated that parity stage of breeding females is an important factor affecting various aspects of reproductive performance in goats. In this respect the present study concurs with many previous ones conducted on the same subject. Mating season, however, was not found to have a statistically significant effect on reproduction traits, although this does not necessarily preclude its biological significance. Numerically, the range of predicted values for traits such as conception and prolificacy rates among mating seasons were quite large, but the variability in within-season environmental conditions among the three production cycles was too high to permit clear expression of statistical differences in these measures. In general, however, maximum reproductive performance in goats kept on semi-arid pastures in northern Kenya can be expected to be achieved when does are mated during the short dry season in January or February. But, at the same time, this management strategy would lead to very large mortality rates in youngstock, so that in terms of the number of kids weaned the best performance would be obtained if mating took place during and towards the end of the long dry season, so that birth and weaning coincides with the long rains and the beginning of the following long dry season, respectively. Differences in kid survival among mating seasons were marked, particularly between the latter mating season group and those in which breeding took place between the months of December and February. The results demonstrated that restricted breeding can be an effective means to manipulate kid mortality. Similar conclusions apply with respect to milk yield, which was an important risk factor affecting kid survival until weaning. Although growth performance of kids until weaning differed markedly among mating seasons, these had largely disappeared by one year of age. Therefore, seasonal breeding does not seem to confer any major advantage in terms of growth performance of youngstock per se. However, account should be taken of the fact that juvenile developmental rates have an impact on survival rates, especially in the early stages of life, and thus affect the expected total liveweight production of youngstock per time unit.

One of the major conclusions that emerged from the herd productivity assessment is that under the current production conditions reproductive performance traits are far less important as contributors to biological productivity than is often assumed. This is due to two main reasons. Firstly, if juvenile mortality is high, resources invested in producing offspring, i.e. in maintaining reproductively active breeding females, are wasted, and therefore youngstock mortality will tend to outweigh reproductive performance traits in its effect on biological productivity. The results of the eigenvector sensitivity analyses carried out on the transition [page 159↓]matrices for each mating season group clearly supported the contention that juvenile survival rate is the most important factor determining overall energetic efficiency. Secondly, in order to assess the contribution of individual traits to, or the effects of changes in, individual traits on herd efficiency their impact on the entire production process, particularly on herd dynamics, must be considered. This can be difficult to achieve with conventional production indices on which most comparative studies have relied, while it is explicitly taken into account by the steady-state herd model proposed in this study. The distortions in rank orders which may emerge from an analysis carried out on the basis of conventional production indices may be substantial.

To summarize, the general impression that emerges from this study is that restricted breeding can effectively be used as a management control to manipulate overall biological herd productivity primarily because of its positive effect on youngstock mortality rates. In contrast, yield levels, i.e., growth and milk performance, are less important as determinants of biological herd productivity, once their effect on youngstock mortality has been accounted for. Ultimately, this is the main reason for the strong positive relationship that was detected between energetic efficiency and asymptotic herd growth rate. Therefore, an interesting suggestion arising from this part of the study is that a simplified productivity ranking of goat herds maintained under similar environmental and management conditions could be carried out on the basis of asymptotic growth rate, for which only stage-specific vital rates for each alternative considered need to be known.

The results of the increased milk offtake scenario re-emphasized the statement that has frequently been made before by ecologists, namely that the combined production of meat and milk for human (subsistence) consumption confers a distinctive advantage over the specialised single-purpose meat production usually found in commercialised extensive livestock production systems. The ranking of mating season groups was little affected by the change in production strategy; rather, in both the baseline and the increased milk offtake scenario, joining does at the peak of the long dry season proved to be the optimal management strategy. However, whether restricted breeding is biologically superior to an aseasonal breeding management, as is often practised by pastoral producers, remains an open question. The results of the productivity assessment for the simulated aseasonal breeding regime seemed to indicate that the potential improvements in biological productivity that could be realized by shifting to a controlled breeding regime are perhaps much smaller than is usually presumed. But, clearly, this conclusion needs to be confirmed by additional empirical evidence. Furthermore, it should be kept in mind that controlled breeding requires a much higher standard of management than those currently being practised by many pastoral producers. This points to the fact that although assessing productivity in biological terms is a necessary step because its components are also important determinants of economic efficiency, it is not sufficient for a final rating of management alternatives. Priority for further research should therefore be placed, firstly, on obtaining more reliable estimates of biological productivity in aseasonally reproducing pastoral goat herds. Secondly, socio-economic studies of pastoral livestock operations are required in order to quantify the economic benefits associated with restricted and unrestricted breeding management. Because of the complex interrelationships between decision making at the herd and at the pastoral household level, such studies should be based on a comprehensive and formal analysis of the objectives of pastoral producers.

Carrying out additional experiments to test the hypothesis that biological productivity of continuously reproducing SEA goat herds is lower than that achieved when breeding is restricted to the long dry season would also offer the opportunity to investigate the effects of changes in age at first breeding and kidding interval on energetic efficiency. Management interventions aimed at reducing both of these variables in goat herds have often been stated to lead to significant improvements in overall biological efficiency, although so far little effort has been put into quantifying these benefits. In order to achieve comparability to the present results, the suggested experiments should, at the very least, take place under similar environmental and management conditions, and the age-at-first-breeding treatment variable should include as a reference level an age of about 15 months, which corresponds approximately to that observed in the present experiment. Based on the results of the sensitivity analyses performed in this work, it may be hypothesised that the effect of kidding interval on energetic efficiency will depend on the age at first breeding and vice versa , such that a two-way-factorial treatment structure appears to be warranted.

With respect to the assessment of biological productivity at the herd level, the methodology developed in the present work is basically an extension of the approach proposed by Baptist (1992b). While the latter assumes that the same inherent survival, fecundity, development rates, as well as yield levels are applied to all breeding and (except, of course, for fecundity rates and milk yields) surplus females alike, the stage-structured population dynamics model employed here is much more flexible since it permits the modelling of life histories in which vital rates, growth and milk performances vary with age, stage, or a combination of both factors. Generally, the appropriate definition of relevant life-cycle stages is entirely dependent on the species and on the production context considered, and is not limited to the parity stages that were used in modelling population dynamics of SEA goat herds. Perhaps most importantly, the present work has demonstrated that assessing biological herd productivity can be formalised in a non-linear programming model which determines the optimal stage-specific culling policy that maximizes the chosen biological or [page 160↓]economic productivity criterion, subject to the constraints that herd size and structure remain constant.

The present work has emphasized the importance of utilizing an optimality approach for obtaining a common basis on which management alternatives can be compared in terms of their effect on energetic efficiency. Many other approaches to steady-sate herd productivity modelling, such as that of James and Carles (1996) and Upton (1989, 1993), are based on estimated offtake rates for a given management alternative or production system and do not attempt to simultaneously optimise herd structure and offtake rates in order to maximize a specific performance criterion. The results of this study have shown that herd productivity is greatly affected by the choice of culling policy and thus herd structure. Of course, herd management affects herd productivity not only through offtake decisions but, if potential herd productivity is to be assessed, at least this major source of variation should be controlled for in comparative assessments. The reason for this is that productivity assessments may be biased if they are based on observed, suboptimal offtake decisions.

The herd productivity assessment procedure developed in this work should be seen as a device with which standardized comparisons of productivity in livestock herds can be carried out. As was illustrated by the results of the sensitivity analyses, steady-state herd productivity models of this type can also be a valuable aid in understanding or optimising the production system looked at. The required input data can either originate from field experiments or livestock surveys, but can as well be generated by simulation models of livestock production systems such as the one described by Bosman et al. (1997b). With regard to simulation experiments, steady state herd productivity modelling can represent an important complementary step in evaluating simulation results. For example, in a simulation experiment, Bosman et al. (1997b: 570) assessed goat herd productivity in terms of RPI and FPI, and argued that productivity indices may be seriously affected by adult liveweight if the amount of some or all feeds available may be limited. This problem occurs mainly because RPI and FPI do not qualify as true measures of biological productivity. Differential effects of adult liveweight on efficiency through variations in the intake above maintenance and, hence, through restrictions in herd size could easily be assessed through steady-state herd productivity modelling with energetic efficiency as the objective criterion. Upon using the approach presented in this work and introducing an additional constraint with respect to total feed energy consumption into the non-linear programming model, one could explicitly take into account differences in mature size (and thus in inputs required and outputs produced) and obtain undistorted estimates of biological productivity at the herd level.


© Die inhaltliche Zusammenstellung und Aufmachung dieser Publikation sowie die elektronische Verarbeitung sind urheberrechtlich geschützt. Jede Verwertung, die nicht ausdrücklich vom Urheberrechtsgesetz zugelassen ist, bedarf der vorherigen Zustimmung. Das gilt insbesondere für die Vervielfältigung, die Bearbeitung und Einspeicherung und Verarbeitung in elektronische Systeme.
DiML DTD Version 3.0Zertifizierter Dokumentenserver
der Humboldt-Universität zu Berlin
HTML generated:
08.10.2004